Hi Roland, you wrote:

As I am still on my spline templates, I came on to the following beast:

template<class T> struct index_vector : public vector<std::pair<T, vector<ublas::vector<T>::size_type> > > {};

This should be an index vector that relates to a double (first) a vector of indices (second) which happens to be different for each 'first'. So far so easy.

I can say:

index_vector<double> IY; IY.resize(5); IY[0].first = 3.14; IY[0].second.reset(vector<double>(3)); IY[0].second[2] = 1;

But I cannot:

index_vector<double> IZ(IY);

or

index_vector<double> IZ; IZ = IY;

or IZ.reset(IY);

This compiles, but gives me runtime assertions because of size mismatches.

Yes. The following mix of std::vector and ublas::vector seems to work for me under GCC 3.2.1: ---------- #include <vector> #include <boost/numeric/ublas/vector.hpp> using namespace std; using namespace boost::numeric; template<class T> struct index_vector : public vector<std::pair<T, ublas::vector<typename ublas::vector<T>::size_type> > > {}; int main () { index_vector<double> IY; IY.resize(5); IY[0].first = 3.14; IY[0].second.reset(ublas::vector<std::size_t>(3)); IY[0].second[2] = 1; index_vector<double> IZ1(IY); index_vector<double> IZ2; IZ2 = IY; } ---------- Do you need ublas algorithms for the outer or the inner vector? Otherwise you could use std::vector instead.

I think my derived class is not 'copy constructible' and 'assignable'. Is there a simple way to fix this?

I believe yes: we could drop the requirement that assignments have to be size conformant. But this requirement is pretty useful when doing basic linear algebra IMHO. So I'd like to hear other opinions...

(I am not even sure whether this is a uBLAS matter. So please simple tell me, and I will try to learn the missing parts.)

It's definitely an ublas matter: should we increase standard compatibility or stay with the size conformance restriction. Best, Joerg